Screening Cardiac Amyloidosis via AI Echocardiography: From Workflow Automation and Radiomics to Navigating the ‘Black Box’

Cardiac Amyloidosis (CA) is often underdiagnosed and diagnosed late because its phenotype overlaps with that of hypertensive cardiovascular disease, hypertrophic cardiomyopathy (HCM), aortic stenosis and heart failure with preserved ejection fraction (HFpEF). Transthoracic echocardiography serves as the primary clinical gateway for screening, yet standard visual interpretation remains limited by operator dependence and variability in recognizing subtle features. Artificial intelligence (AI) is transforming this pathway through two principal paradigms: the automation of image acquisition, view classification, and deformation imaging (e.g., global longitudinal strain), as well as end-to-end deep video learning alongside radiomic texture analysis. Recent multicentre validations demonstrate that video-based deep neural networks can detect CA with high accuracy in external cohorts, while feature-based and radiomic approaches aim to capture subtle, spatial structural patterns that may escape routine visual assessment, offering complementary quantitative metrics. While spatial heatmaps such as Grad-CAM can provide good visual intuition about what parts of the image a model used to make a classification decision, they do not provide any true insight into the underlying pathophysiology. This narrative review assesses the current clinical applications of AI in echocardiography for the screening of cardiac amyloidosis. Automated workflows, deep video architectures, and the use of radiomics are presented, along with several different explainable AI (XAI) frameworks. The review also outlines the challenges to the safe clinical application of AI in echocardiography from a regulatory, ethical, and governance perspective.